LimJianYang
Project

J.A.R.V.I.S.

#tars#iron man#jarvis#agi#ai#assistant

I want to build J.A.R.V.I.S.

Y’know, J.A.R.V.I.S.—Tony Stark’s personal AI assistant from the film Iron Man? J.A.R.V.I.S., or I should just call Jarvis from now, is an acronym for “Just A Rather Very Intelligent System”. It is a formless, highly intelligent computer system that assists Tony Stark in his day-to-day activities. Jarvis is capable of speech, understanding natural language, planning, suggesting solutions, has personality, and access points to interact with the physical world.

A similar concept could be TARS from Interstellar. TARS’ most distinctive trait is its personality, which is a blend of humour, sarcasm, and honesty.

Before ChatGPT, building an AI persoanl assistant seemed like a rather challenging task. Back then, interpretation of natural language was only mid at best. Now with ChatGPT, we have an overpowered autocomplete tool that can generate consistent and coherent responses that’s capable of faking understanding, convincingly.

“Fake it till you make it”? How can a conman act smart without actually making some smart comments? Similarly, how can ChatGPT fake understanding and intelligence so well without having the slightest bit of potential for “true intelligence”?

I believe that we are on the right track with ChatGPT, Gemini, Claude, and many other frontier LLMs towards our own personal assistant. Understanding a command is the first step to getting it done.

Martin Skow Røed, CTO of Databutton once told me during my internship interview (something along the lines of) “after all, there’s definitely a prompt to every right answer”. We have the tools to create such an AI personal assistant.

We simply have to ask the right questions.


Understanding the Situation

1. Seek inspiration from actual secretaries

2. Hierarchical agents

3. System 1 and System 2

4. Brain with hands


An all-purpose, all-intelligent personal AI assistant is too ambitious and too big of a project to take on at once. Like our assistant, we too have to break it down into smaller, more manageable tasks.

Let’s begin with building a personal research assistant to make sound investing decisions. We’ll call this Jarvis-I.

Jarvis-I will be able to:

  1. Learn what the user looks out for when analysing financial statements.
  2. Learn the user’s investment philosophy.
  3. Cover the user’s blind spots, neutralising the user’s biases.
  4. Ask insightful questions and discuss with the user on the soundness of the investment idea.
  5. Create graphics to help the user visualise key information (making it more human-friendly).
  6. Take notes about key points and finetune the thesis along the way.

To be continued.

← Prev: Artificial Gene Back to Blog → Next: On-Device Retri